arX

iv:1

412.

3840

v1 [

astr

o-ph

.SR

] 1

1 D

ec 2

014

Mon. Not. R. Astron. Soc. 000, 1–20 (2014) Printed 13 June 2018 (MN LATEX style file v2.2)

The UK Infrared Telescope M33 monitoring project. IV.

Variable red giant stars across the galactic disc

Atefeh Javadi1, Maryam Saberi3, Jacco Th. van Loon2, Habib Khosroshahi1,

Najmeh Golabatooni3, and Mohammad Taghi Mirtorabi31School of Astronomy, Institute for Research in Fundamental Sciences (IPM), P.O. Box 19395-5531, Tehran, Iran2Astrophysics Group, Lennard-Jones Laboratories, Keele University, Staffordshire ST5 5BG, UK3Physics Department, Alzahra University, Vanak, 1993891176, Tehran, Iran

Resubmitted: 2014

ABSTRACT

We have conducted a near-infrared monitoring campaign at the UK InfraRed Tele-scope (UKIRT), of the Local Group spiral galaxy M33 (Triangulum). The main aimwas to identify stars in the very final stage of their evolution, and for which the lumi-nosity is more directly related to the birth mass than the more numerous less-evolvedgiant stars that continue to increase in luminosity. In this fourth paper of the series,we present a search for variable red giant stars in an almost square degree region com-prising most of the galaxy’s disc, carried out with the WFCAM instrument in the Kband. These data, taken during the period 2005–2007, were complemented by J- andH-band images. Photometry was obtained for 403 734 stars in this region; of these,4643 stars were found to be variable, most of which are Asymptotic Giant Branch(AGB) stars. The variable stars are concentrated towards the centre of M33, more sothan low-mass, less-evolved red giants. Our data were matched to optical cataloguesof variable stars and carbon stars and to mid-infrared photometry from the SpitzerSpace Telescope. Most dusty AGB stars had not been previously identified in opticalvariability surveys, and our survey is also more complete for these types of stars thanthe Spitzer survey. The photometric catalogue is made publicly available at the Centrede Donnees astronomiques de Strasbourg.

Key words: stars: evolution – stars: luminosity function, mass function – stars:mass-loss – stars: oscillations – galaxies: individual: M33 – galaxies: stellar content

1 INTRODUCTION

The Local Group galaxy Triangulum (Hodierna 1654) –hereafter referred to as M33 (Messier 1771) – offers us aunique opportunity to study stellar populations, their his-tory and their feedback across an entire spiral galaxy and inparticular in its central regions, that in our own Milky Wayare heavily obscured by the intervening dusty disc (van Loonet al. 2003; Benjamin et al. 2005). Our viewing angle withrespect to the M33 disc is more favourable (56–57◦ – Zarit-sky, Elston & Hill 1989; Deul & van der Hulst 1987) thanthat of the larger M31 (Andromeda), whilst the distance toM33 is not much different from that to M31 (µ = 24.9 mag– Bonanos et al. 2006). For these reasons, numerous surveyshave been conducted to study M33 in different wavelengthranges including optical (Macri et al. 2001; Hartman et al.2006; Massey et al. 2006); radio (Engargiola et al. 2003;Rosolowski et al. 2007); X-ray (Pietsch et al. 2004); andinfrared (IR – Two Micron All Sky Survey [2MASS], Skrut-skie et al. 2006; McQuinn et al. 2007). Large populations of

Asymptotic Giant Branch (AGB) stars have been identifiedin M33 (Cioni et al. 2008), as well as red supergiants (RSGs)up to progenitor masses in excess of 20 M⊙ (Drout, Massey& Meynet 2012). Many of them are dusty Long Period Vari-ables (LPVs – McQuinn et al. 2007; Thompson et al. 2009),and these have been found also in the central parts of M33(Javadi, van Loon & Mirtorabi 2011a; Javadi et al. 2013).

In the final stage of stellar evolution, low–medium mass(0.8–8 M⊙) stars enter the AGB phase (Marigo et al. 2008)and more massive stars (M > 8 M⊙) enter the RSG phase(Levesque et al. 2005; Levesque 2010). These two phases ofevolution trace stellar populations over a range in age from∼ 10 Myr to ∼ 10 Gyr, and hence the evolution of their hostgalaxies over essential all cosmological time. Radial pulsa-tions of the atmospheric layers in AGB stars and RSGs yieldlong-period variability of order 150–1500 days in the pho-tometric light curves (e.g., Wood et al. 1992; Wood 1998;Pierce et al. 2000; Whitelock et al. 2003). Most evolved AGBstars pulsate in the fundamental mode, while less evolvedAGB stars and RSGs pulsate in an over-tone; as a result

c© 2014 RAS

2 Javadi et al.

the amplitude of variability expressed in magnitude of AGBstars are larger than that of RSGs and less evolved AGBstars. These LPVs are powerful tools to study the star for-mation history of galaxies, and to this aim various variabil-ity surveys have been conducted of M 33 over recent years(Macri et al. 2001; Mochejska et al. 2001a,b; Hartman et al.2006; Sarajedini et al. 2006; McQuinn et al. 2007).

The coolest (T ∼ 3000–4000 K) and most luminous(∼ 10 000–60 000 L⊙) AGB stars create large amounts ofdust in the stars atmosphere. This dust is ejected into spaceand can cloak the star, especially in the optical light, andadd luminosity at IR wavelengths. Likewise, RSGs standout especially at IR wavelengths. Hence, among all sur-veys, those with the Spitzer Space Telescope (McQuinn etal. 2007) and the UK InfraRed Telescope (UKIRT, Javadiet al. 2011a) were more successful in detecting dusty LPVs.Since these evolved stars shed a large amount of mass intothe interstellar medium (ISM), they are important factorsin changing the chemical composition of galaxies and incre-menting the rate of stellar birth.

The main objectives of our project are described inJavadi, van Loon & Mirtorabi (2011c): to construct the massfunction of LPVs and derive from this the star formation his-tory in M33; to correlate spatial distributions of the LPVsof different mass with galactic structures (spheroid, disc andspiral arm components); to measure the rate at which dust isproduced and fed into the ISM; to establish correlations be-tween the dust production rate, luminosity, and amplitudeof an LPV; and to compare the in situ dust replenishmentwith the amount of pre-existing dust. Paper I in the seriespresented the photometric catalogue of stars in the innersquare kpc (Javadi et al. 2011a), with Paper II presentingthe galactic structure and star formation history (Javadi,van Loon & Mirtorabi 2011b), and Paper III presenting themass-loss mechanism and dust production rate (Javadi etal. 2013). This paper describes the extension of the surveyto a nearly square degree area covering much of the M 33optical disc. Subsequent papers in the series will cover thestar formation history and mass return in this enlarged area.

In Section 2 we present the observational data, methodof photometry on the images, and accuracy of these measure-ments. The methodology and completeness of our search forLPVs, and characterization of their amplitude of variabilityare discussed in Section 3. Section 4 describes the photo-metric catalogue of all detected stars, which is made avail-able at the Centre de Donnees astronomiques de Strasbourg(CDS). In section 5 we describe the properties and distribu-tion of the detected LPVs and also compare these with other(optical and IR) variability surveys and stellar catalogues.Section 6 summarizes and concludes the results.

2 OBSERVATIONS

Observations were made with three of UKIRT’s imagers:UIST, UFTI and WFCAM. UIST and UFTI cover the cen-tral part (≈ 1 kpc2) and the photometry, star formationhistory and mass return in this region was explained in Pa-pers I, II and III. In this sequel we focus on the data fromWFCAM, which cover a much larger part of M33.

Table 1. Log of our observations of each of four tiles (“Q”).

Date (y m d) Q Filter Epoch tint (min) Airmass

2005 09 18 3 K 1 20.3 1.035–1.0582005 09 18 2 K 1 20.3 1.072–1.1102005 09 18 4 K 1 20.3 1.248–1.3382005 09 19 1 K 1 20.3 1.021–1.0182005 10 18 3 K 2 20.3 1.019–1.0212005 10 18 2 K 2 20.3 1.025–1.0402005 10 18 4 K 2 20.3 1.053–1.0832005 10 18 1 K 2 20.3 1.101–1.1492005 11 04 1 K 3 20.3 1.018–1.0232005 11 04 2 K 3 13.5 1.028–1.0362005 12 23 2 K 4 27.0 1.019–1.0222005 12 23 3 K 3 20.3 1.028–1.0462006 07 21 1 K 4 20.3 1.425–1.3252006 07 21 2 K 5 20.3 1.287–1.2142006 07 21 3 K 4 20.3 1.183–1.1322006 07 21 4 K 3 20.3 1.109–1.0742006 10 28 1 K 5 27.0 1.294–1.1262006 10 28 1 J 1 20.3 1.102–1.0762006 10 29 1 K 6 20.3 1.445–1.3472006 10 29 1 H 1 27.0 1.295–1.2092006 10 29 1 J 2 27.0 1.115–1.0622006 10 30 1 J 2 33.8 1.200–1.1092006 10 30 4 K 4 33.8 1.085–1.0442006 10 31 4 H 1 6.8 1.301–1.3012006 12 05 2 K 6 20.3 1.019–1.0252006 12 12 3 K 5 20.3 1.082–1.0522006 12 12 3 H 1 20.3 1.040–1.0272007 01 14 1 K 7 20.3 1.124–1.1692007 01 14 1 J 3 20.3 1.217–1.2842007 01 14 1 H 2 20.3 1.342–1.4412007 01 15 2 K 7 20.3 1.119–1.1632007 01 16 2 H 1 20.3 1.063–1.0922007 01 17 3 J 1 20.3 1.031–1.0472007 01 18 2 J 2 20.3 1.029–1.0442007 01 25 3 K 6 20.3 1.072–1.1042007 01 25 3 H 2 20.3 1.161–1.2152007 09 14 1 K 8 20.3 1.122–1.0862007 09 14 1 J 4 13.5 1.070–1.058

2007 09 14 1 H 3 13.5 1.046–1.0382007 09 19 2 K 8 20.3 1.774–1.6062007 10 04 2 J 3 13.5 1.208–1.1812007 10 04 2 H 2 13.5 1.155–1.1322007 10 13 3 K 7 20.3 1.108–1.0762007 10 13 3 H 3 13.5 1.056–1.0462007 10 24 4 K 5 20.3 1.135–1.0972007 10 24 3 J 2 13.5 1.078–1.0642007 10 24 4 J 1 13.5 1.050–1.0412007 10 24 4 H 2 13.5 1.025–1.022

2.1 WFCAM

The monitoring campaign comprises observations with theWide Field CAMera (WFCAM) such that four separatelypointed observations (“tiles”, viz. M33-1, M 33-2, M33-3and M33-4) may be combined to cover a filled square areaof sky covering 0.89 square degree (13 kpc × 13 kpc) ata pixel size of 0.′′4. The approximate centres of the cam-era pointings are respectively (1h33m19.s30, +30◦32′50′′),(1h34m22.s50, +30◦32′50′′), (1h34m22.s50, +30◦46′23′′),(1h33m19.s30, +30◦46′23′′). Observations were made in theK band (UKIRT filter K98) over the period September 2005–October 2007 (Table 1). On some occasions observations

c© 2014 RAS, MNRAS 000, 1–20

M33 monitoring. IV 3

were made also in the J band and/or H band (UKIRT filtersJ98 and H98, respectively) to provide colour information.The total integration on one tile was achieved through fourseparate exposures. The number of epochs varies per tile be-tween five and eight, but overlapping regions will have beenobserved more frequently.

The large pixel scale of WFCAM means that the pointspread function (PSF) is not always adequately sampled forthe crowded fields in M33 under modal observing condi-tions; to remedy this we employed a 3 × 3 microsteppingscheme to improve the sampling of the PSF. At each positionof the nine-point microstepping sequence a 5-sec exposurewas taken, and a small offset (“jitter”, ≈ 0.′′5) was appliedbetween each of nine subsequent microstep sequences. Thiscycle was repeated three times. Thus, a typical observationaccrued a total integration time of ≈ 20 min; however, inpractice fewer or more repeats were performed dependingon conditions, time available, or for technical reasons (inter-ruptions).

The images were processed using the WFCAM pipelineby the Cambridge Astronomy Survey Unit (CASU –http://casu.ast.cam.ac.uk/). The main steps include:

• Dark current correction;• Flat field correction to remove pixel sensitivity differ-

ences and gain differences between data channels and be-tween detectors;

• Confidence map generation: a normalized inverse weightmap denoting the confidence associated with the flux valuesin each pixel;

• Defringing, provided the fringe spatial pattern wasavailable;

• Sky subtraction, either using the ditter sequences them-selves (our case) or using observations of offset sky regions;

• Image persistence and detector crosstalk correction, i.e.modelling and removing electronic effects;

• Combine (“interleave”) the images from the microstep-ping sequence;

• Shift and average individual images from the jitteringsequence;

• Catalogue generation. Objects are identified from theimages and listed in FITS binary tables. This catalogue in-cludes assorted aperture flux measures, intensity-weightedcentroid estimates, and shape information such as intensity-weighted second moments to encode the equivalent ellipticalGaussian light distribution;

• Astrometric calibration. All objects in the catalogue arematched to astrometric standards to define a World Coor-dinate System (WCS) for each image/catalogue;

• Photometric zeropoint from comparison of instrumentalmagnitudes with 2MASS (K band is Ks).

Our programme IDs are U/05B/18, U/06B/40 andU/07B/17. We have complemented our data with WFCAMarchival data taken for two projects, viz. U/05B/7 andU/05B/H47 which we briefly describe below.

2.1.1 U/05B/7

The data of this programme (Cioni et al. 2008) were taken tosurvey the luminous red giant stars of Local Group galaxies.M33 was observed on four occasions: 29 and 30 September,24 October, 5 November and 16 December 2005 (Table 2).

Table 2. Log of WFCAM observations of each of four tiles (“Q”),from programme U/05B/7.

Date (y m d) Q Filter Epoch tint (min) Airmass

2005 09 29 4–1 J 1 1.0 1.196–1.1972005 09 29 4–1 H 1 4.5 1.206–1.2272005 09 29 4–1 K 1 4.5 1.227–1.0232005 09 30 2–1 J 1 1.0 1.023–1.0232005 09 30 2–1 H 1 4.5 1.024–1.0242005 09 30 2–1 K 1 4.5 1.028–1.0282005 09 30 2–3 J 1 1.0 1.055–1.0562005 09 30 2–3 H 1 4.5 1.060–1.0602005 09 30 2–3 K 1 4.5 1.069–1.0692005 10 24 1–1 J 1 1.0 1.335–1.3332005 10 24 1–1 H 1 4.5 1.320–1.3192005 10 24 1–1 K 1 4.5 1.293–1.2922005 10 24 1–2 J 1 1.0 1.050–1.0202005 10 24 1–2 H 1 4.5 1.340–1.2102005 10 24 1–2 K 1 4.5 1.210–1.1202005 10 24 1–3 J 1 1.0 1.204–1.2022005 10 24 1–3 H 1 4.5 1.193–1.1932005 10 24 1–3 K 1 4.5 1.175–1.1742005 10 24 1–4 J 1 1.0 1.158–1.1572005 10 24 1–4 H 1 4.5 1.145–1.1442005 10 24 1–4 H 1 4.5 1.130–1.1302005 11 05 3–1 K 1 4.5 1.073–1.0732005 11 05 3–1 H 1 4.5 1.064–1.0642005 11 05 3–1 J 1 1.0 1.055–1.0552005 11 05 3–2 K 1 4.5 1.052–1.0522005 11 05 3–2 H 1 4.5 1.046–1.0462005 11 05 3–2 J 1 1.0 1.039–1.0382005 11 27 1–1 J 2 1.0 1.292–1.2952005 12 16 1–1 J 2 1.0 1.107–1.1062005 12 16 1–1 H 2 4.5 1.101–1.1012005 12 16 1–1 K 2 4.5 1.090–1.0892005 12 16 1–2 J 2 1.0 1.079–1.0782005 12 16 1–2 H 2 4.5 1.074–1.0742005 12 16 1–2 K 2 4.5 1.065–1.0642005 12 16 1–3 J 2 1.0 1.055–1.0542005 12 16 1–3 H 2 4.5 1.051–1.0512005 12 16 1–3 K 2 4.5 1.044–1.044

2005 12 16 1–4 J 2 1.0 1.038–1.0382005 12 16 1–4 H 2 4.5 1.035–1.0352005 12 16 1–4 K 2 4.5 1.031–1.031

JHKs photometry was obtained from a mosaic of four fields(instead of the one central field in our case), covering an area≈ 3 square degrees. The H and Ks data were acquired em-ploying a three-point jitter pattern with 3×3 microsteppingand 10-sec exposures per position, giving a total integrationtime of 270 sec. The J data were acquired using a five-pointjitter pattern with three 10-sec exposures but no microstep-ping, resulting in a total integration time of 150 sec. Thedata were processed by the CASU.

2.1.2 U/05B/H47

The M33 data for this programme (PI: M. Irwin) were ac-quired on 20 October 2005 (Table 3). The covering area isnearly identical to ours, with the camera pointings calledM33-position 1, M33-position 2, M33-position 3 and M33-position 4, respectively.

c© 2014 RAS, MNRAS 000, 1–20

4 Javadi et al.

Figure 1. Combined WFCAM K-band mosaic of M 33. The previously investigated, square-kpc area is delineated with a box.

Table 3. Log of WFCAM observations of each of four tiles (“Q”),from programme U/05B/H47.

Date (y m d) Q Filter Epoch tint (min) Airmass

2005 10 20 1 K 1 8.3 1.077–1.0592005 10 20 2 K 1 8.3 1.052–1.0412005 10 20 3 K 1 8.3 1.038–1.030

2005 10 20 4 K 1 8.3 1.028–1.023

2.2 Images

We present the combined, square-degree mosaic of M33in the K band in figure 1. The previously investigated,square-kpc area is delineated with a box. While the spi-ral structure is evident the images are much less affectedby extinction by interstellar dust than images at opticalwavelengths. The very bright Galactic foreground red gi-ant HD9687 (1h35m26s, +30◦46.′5) has left a trail of ghostimprints to its East, but saturation is not normally a prob-lem across the mosaic. Most stars above the tip of the redgiant branch (RGB) are resolved also with WFCAM, exceptwithin the central few arcsec dominated by the nuclear starcluster.

c© 2014 RAS, MNRAS 000, 1–20

M33 monitoring. IV 5

2.3 Cross correlation of catalogues

The photometric catalogues of M33 were retrieved from thepublic WFCAM Science Archive (WSA). Only the overlap-ping area of the other catalogues (and images) with thosefrom our own programme will be considered here; they aremeant to provide further epochs that increase the sensitivityand reliability of the variability detection.

The FITS catalogues contain only un-calibrated fluxes(f , in counts) obtained in a series of apertures. By knowingthe photometric zeropoint at unity airmass, m0, the extinc-tion coefficient, a (Krisciunas et al. 1987), the airmass atthe start and end of observation, zstart and zend, and ex-posure time, t, the (telluric) extinction-corrected and flux-calibrated magnitudes, m, are determined by:

m = m0 − 2.5 log(

f

t

)

− a×(

zstart + zend2

− 1)

. (1)

In order to determine the mean magnitude and lightcurve of all stars over the epochs in which they were de-tected, unique IDs need to be assigned to all individual stars.To this aim, we cross-identified each star within every cat-alogue. The matches were obtained by performing searchiterations using growing search radii, in steps of 0.′′1 outto 1′′, on a first-encountered first-associated basis but afterordering the principal photometry in order of diminishingbrightness (to avoid rare bright stars being erroneously as-sociated with any of the much larger number of faint stars).

2.4 Photometric calibration

Given the high level of crowding especially towards the cen-tral parts of M33, and the importance of accurate relativephotometry between epochs in order to correctly separatevariable from non-variable sources, it is essential to check theaccuracy of the WSA catalogues. To this aim, we performedPSF photometry using the DAOPhot package within IRAF(Stetson 1987) on one of the frames (from camera 1) in thecentral part field M33-3. An empirical constant PSF modelwith a 2D elliptical Gaussian function was used to constructthe PSF from seven isolated stars. Then, the allstar taskwas used to estimate the instrumental magnitude for 32,627stars identified with the daofind task. The transformationfactor from instrumental magnitude to standard magnitudewas obtained from the standard magnitudes of 30 stars incommon from the UIST catalogue (Paper I), hence derivingm0 for the WFCAM data.

The celestial coordinates of the stars were calculated us-ing the IRAF ccmap–cctran tasks. In this frame, the WSAcatalogue lists 36 301 stars. We cross-identified these withtheDAOPhot catalogue within a search radius of 1′′; hence,24 140 stars were identified in common. Figure 2 shows thedifference in magnitude between the WSA and DAOPhot

photometry, against the WSA magnitude; the histogram ofmagnitude differences is shown in figure 3. The vast major-ity of stars have magnitudes that are consistent between thetwo methods of measurement within a few tenths of a mag-nitude; among the 3.5 per cent of stars with ∆m > 1 mag,78 per cent are located near (within 5′ of) the centre of M 33where crowding is more severe than elsewhere in the mosaic.This renders just 0.9 per cent of stars with suspect photom-etry. Upon visual inspection of the image, it was found that

Figure 2. Magnitude differences between WSA (aperture) andDAOPhot (PSF) photometry, plotted against WSA magnitude.Stars for which ∆m > 1 mag have been identified based onwhether they are located within the central 5′ or further awayfrom the unresolved centre of M 33.

Figure 3. Histogram of the magnitude differences between WSA(aperture) and DAOPhot (PSF) photometry.

in most cases there is a faint star near a bright star, andthe photometric difference is simply the result of erroneouscross-identification. Consequently, the accuracy of the WSAmagnitudes is acceptable.

c© 2014 RAS, MNRAS 000, 1–20

6 Javadi et al.

Figure 4. Relative calibration values added to the magnitudesin each of the individual frames, as a function of time, over theduration of the UIST and WFCAM monitoring campaigns.

2.4.1 Relative calibration

The relative calibration between frames was obtained fromthe mean magnitudes of ≈ 1000 stars in common within themagnitude interval K ∈ [16...18]. While these will includesome variable stars the vast majority will not vary by morethan 10 per cent rendering their mean magnitude accurateto well within a per cent. Then the photometry from the dif-ferent frames was brought in line with each other by apply-ing corrections that equalised these mean magnitudes. Thecorrections are shown in figure 4, also for the UIST survey(they were not shown in Paper I). Note that the correctionsare small – generally a few per cent and always less than 10per cent.

3 VARIABILITY ANALYSIS

The search for variable stars was done by calculating thevariability index for each star. This index was introducedby Welsh & Stetson (1993) and developed further by Stet-son (1996). In this method, first the observations are pairedon the basis of timespan between observations such thatthe observations of each pair have a timespan less than theshortest period expected for the kind of variable star of in-terest. In case more than two observations were performedwithin the timespan of shortest periodicity, those sets of ob-servations would be paired in more than one pair. Hence,the J index is calculated:

J =ΣN

k=1wk sign(Pk)√

|Pk|

ΣNk=1

wk

. (2)

Here, observations i and j have been paired and for each paira weight wk is assigned; the product of normalized residuals,

Figure 5. Variability index L vs. K-band magnitude.

Pk = (δiδj)k1, where δi = (mi − 〈m〉)/ǫi), is the residual of

measurement i from the mean magnitude, normalized by theerror of the measurement, ǫi; and N is the total number ofobservations. Note that δi and δj may refer to observationstaken in different filters. The J index for non-variable starsis approximately zero as the residuals arising from randomnoise are uncorrelated and their product will therefore tendto zero for large sets of measurements.

The effect of a small number of observations or corruptdata can be limited by means of a backup index, viz. theKurtosis index:

K =1

NΣn

i=1|δi|√

1

NΣN

i=1δ2i

. (3)

The shape of the light variations define the value of theKurtosis index; for example, K = 0.798 for a Gaussian dis-tribution which is concentrated towards the average bright-ness level (as would be random noise), and K → 0 for dataaffected by a single outlier (when N → ∞).

Here, we use the variability index L (Stetson 1996):

L =J ×K

0.798

Σw

wall

, (4)

where Σw is the total weight assigned to a given star andwall is the total weight a star would have if observed in everysingle observation.

Figure 5 shows how the variability index L varies withK-band magnitude. For comparison with our previous UIST

1 Following Stetson (1996), Pk = δ2 − 1 if i = j.

c© 2014 RAS, MNRAS 000, 1–20

M33 monitoring. IV 7

Figure 6. Variability index L vs. K-band magnitude (from theWFCAM data) for the central square kpc field that was monitoredwith UIST (Paper I). The green points are the UIST variable starsthat were detected with WFCAM.

Figure 7. Variability index L histograms for several magnitudesbins in the range Ks = 16–19 mag. Red lines trace the Gaussianfunction fitted to each histogram.

survey (Paper I), in figure 6 we show the distribution for thecentral square kpc of M 33, where we indicate the WFCAMdetections of stars that had been identified as variable inthe UIST survey. Both graphs reveal a noticeable “branch”of stars with larger than usual L between K ∼ 16–18 mag;these are likely AGB stars with Mira-type variability.

Several tests were performed to select the optimal vari-ability index threshold. First we inspected the histograms of

Figure 8. As Fig. 7, but limited to the central square kpc ofM 33. The vertical dotted line marks the limit which is used forselection of variable stars in this part of M 33.

Figure 9. As Fig. 8, but now excluding the central square kpc.

variability index within several magnitude bins in the range16–19 mag for all detected stars including the M33 disc andcentral regions (Fig. 7). To determine the variability indexthreshold, a Gaussian function was fitted to each of thesehistograms. The Gaussian function should be near-perfectlyfitted to those data for low values of L, while it departs fromthe histograms for larger values of L. This appears to hap-pen around L ∼ 0.8 but at smaller values for faint stars;we thus decided, in first instance, to set the threshold atL > 0.7 for the detection of variability.

c© 2014 RAS, MNRAS 000, 1–20

8 Javadi et al.

Figure 10. Near-IR colour–magnitude diagram of M 33 with in-dicated the putative variable stars according to three choices ofvariability index L threshold, for (Left:) the central square kpcand (Right:) the disc.

Next, we test our procedure by comparing the WFCAMselected variables, with L > 0.7, with the UIST catalogue ofvariables in the central square kpc (from Paper I). The per-centage of WFCAM variables within the magnitude rangesshown in figure 7 is about twice that derived from the UISTsurvey. If anything, the opposite would be expected as theUIST survey had a superior cadence of observations. Indeed,when selecting only stars from this central region of M33,the histograms of variability index broaden, and a more ap-propriate threshold would be L > 1.5 (Fig. 8). On the otherhand, a threshold of L > 0.7 seems appropriate for the discof M 33, i.e. excluding the central square kpc (Fig. 9).

To further assess the validity of the choice of variabilityindex threshold, we examined the location of the putativevariable stars in a colour–magnitude diagram (CMD), fordifferent choices of L threshold: L > 0.7, L > 1 and L > 1.5(Fig. 10). While there is no clear difference in the selectionof variable stars on the AGB and RSG portions of the CMD(roughly at Ks = 14–18.5 mag and (J − Ls) > 0.8 mag), allbut the L > 1.5 thresholds result in many putative variablestars at fainter magnitudes and along the blue and brightvertical sequence (around (J − Ks) ∼ 0.5 mag) where nolarge-amplitude red giant variables are expected.

Finally, we inspected the radial distribution of puta-tive variable stars for the case where we apply differentchoices for the L threshold for the central square kpc andthe (remainder of the) disc, as compared to applying a sin-gle threshold for all (Fig. 11). The former case results in a

Figure 11. Radial distribution of putative variable stars where(Left:) different variability index L thresholds were applied for thecentral square kpc and the disc and (Right:) the same thresholdswere applied.

break in gradient around r ∼ 500′′ – i.e. well outside thecentral square kpc – where L > 0.7 results in a seeminglydisproportionally increased number of variable stars closerto the centre. This gradient is not sustained within the cen-tral r ∼ 150′′ – i.e. corresponding roughly to the centralsquare kpc. Indeed, the distribution turns over and muchfewer variable stars are identified once L > 1.5 was applied.The same artificial behaviour is not seen when the same Lthreshold is applied across the entire M33 galaxy.

Based on the histograms, CMD and radial distributionswe decided to apply a single variability threshold of L > 1.5.

3.1 Comparison between the WFCAM and UIST

catalogues of variable stars within the central

square kpc

One final assessment of the variability detection success ismade by a more careful comparison of the common areaof the WFCAM and UIST surveys. After sorting both cata-logues in order of diminishing brightness in K band, matcheswere obtained through successive search iterations using in-creasing search radii, in steps of 0.′′1 out to 1′′. Within thecoverage of UIST (Paper I), using WFCAM we detected11 114 stars; from 18 398 stars detected with UIST, usingWFCAM we detected 10 095 stars. Applying L > 1.5, wefind 667 variable stars located within the central square kpccovered by UIST (out of 4643 in total across the WFCAMcoverage); applying L > 0.7, this becomes 2696 – i.e. morethan three times as many as were found in the more capableUIST survey. Again applying L > 1.5, only 192 out of the812 UIST LPVs were identified with WFCAM (a successrate of 24%). This also suggests that the UIST survey, too,is incomplete.

Figures 12 and 13 show CMDs in which are highlighted

c© 2014 RAS, MNRAS 000, 1–20

M33 monitoring. IV 9

Figure 12. Near-IR CMD of the central square kpc of M 33,showing those stars from the UKIRT/WFCAM survey that wereand were not detected in the UKIRT/UIST survey (Paper I). Avariability threshold of L > 1.5 was applied to select variablestars from the WFCAM survey.

Figure 13. Same as Fig. 12, but for a threshold of L > 0.7.

those variable stars which have and those which have notbeen detected with either WFCAM of UIST, when apply-ing a WFCAM variability index threshold of L > 1.5 andL > 0.7, respectively. These CMDs suggest that L > 1.5is a more sensible threshold than L > 0.7, for two reasons;firstly, it yields a somewhat smaller number of variables withWFCAM than with UIST, which is expected as WFCAMhas more difficulty in isolating blended stars and also theWFCAM survey is based on fewer epochs. Secondly, the

Figure 14. Radial profiles of total stellar density (divided byfour for ease of comparison) and density of variable stars foundin the WFCAM and/or UIST surveys, applying a variability indexthreshold of L > 0.7 (Left) and L > 1.5 (Right), respectively.

Figure 15. Radial profile of the fraction of UIST variables thatis recovered in the WFCAM variability survey.

fraction of UIST variables that were found with WFCAMis smaller but the fraction of WFCAM variables that werefound with UIST is much larger (31%) than when applyingL > 0.7 (14%).

Since the central part of M33 comprises a considerablevariation in stellar density, the same data displayed in fig-ures 12 and 13 are shown in figure 14 as a function of radialdistance from the centre of M 33. The rate at which vari-able stars are detected increases towards the centre, and asa consequence the number of variable stars found in one

c© 2014 RAS, MNRAS 000, 1–20

10 Javadi et al.

Figure 16. Recovery rate of UIST variable stars within the WF-CAM survey, and vice versa, as a function of WFCAM variabilityindex threshold (L).

of the surveys that is not recovered within the other surveyalso increases towards the centre. However, when applying avariability index threshold of LWFCAM > 0.7, the WFCAMsurvey yields many more variable stars than the UIST sur-vey, whilst for LWFCAM > 1.5 this is about equal. This againfavours the choice of the latter threshold. The success rateof the WFCAM survey to recover UIST variable stars dis-plays only a shallow gradient with radial distance from thecentre of M33 (Fig. 15); it averages ∼ 22% for this centralregion (but is ∼ 24% for the square area encompassing thiscircular area).

The recovery rate of UIST variables within the WF-CAM catalogue of variable stars, and vice versa, is plottedas a function of WFCAM variability index threshold in fig-ure 16. At LWFCAM > 1.5, a near-equal fraction of variablestars is recovered within each of the surveys (∼ 30%) – theUIST survey is slightly more successful than the WFCAMsurvey, as expected (see above). As LWFCAM increases, thenumber of variable stars identified in the WFCAM surveydecreases but these variable stars will be the ones with largeramplitudes. Hence, the UIST survey will be more completein including those WFCAM variable stars, reaching ∼ 90%success rate for LWFCAM > 5. On the other hand, the WF-CAM survey will miss more of the UIST variable stars (gen-erally the ones with smaller amplitudes), and its successrate drops to just one per cent for LWFCAM > 5. Hence, thechoice for LWFCAM > 1.5 is supported once again.

3.2 Amplitudes of variability

We estimate the amplitude of variability by assuming a si-nusoidal lightcurve shape. The amplitude is then:

A ≈ 2× σ/0.701 (5)

Figure 17. Estimated amplitude, AK, of variability vs. K-bandmagnitude. Stars with ≤ 6 measurements are highlighted in red.

where σ is the standard deviation in our data and 0.701 isthe standard deviation of a unit sine function. The standarddeviation is uncertain when the number of measurements(N) is small; we showed in Paper I that N = 6 is the mini-mum acceptable to reach ∼ 10% fidelity.

The estimated K-band amplitude is plotted vs. K-bandmagnitude in figure 17. As is well known by now (Wood etal. 1992; Wood 1998; Whitelock et al. 2003; Paper I), theamplitude of variability exhibits a clear tendency to dimin-ish with increasing brightness, which is partly due to thedefinition of magnitude as a relative measure rather thanless powerful pulsation (van Loon et al. 2008). The ampli-tude is generally about a magnitude or less, but a smallfraction of variable stars (8%) seem to reach AK > 2 mag.Very dusty AGB stars – which are rare – are known to reachsuch large amplitudes (Wood et al. 1992; Wood 1998; White-lock et al. 2003). Among these extreme variables, 311 starshave N ≤ 6; excluding these stars, only 20 stars remain withAK > 3 mag. The brightest variables, with Ks < 13 mag,are foreground red giants; among the faintest stars, withKs > 17 mag, our survey becomes increasingly less sensitiveto small-amplitude variables.

4 DESCRIPTION OF THE CATALOGUE

The photometric catalogue including all variable and non-variable stars is made publicly available at the Centre deDonnees astronomiques de Strasbourg (CDS). The contentis described in Table 4. It is composed of two parts, partI comprising the mean properties of the stars and part IItabulating all the photometry (for the benefit of generatinglightcurves, for instance).

The astrometric accuracy of the catalogue is ≈ 0.2′′

r.m.s., tied to the 2MASS system. This accuracy was found

c© 2014 RAS, MNRAS 000, 1–20

M33 monitoring. IV 11

Table 4. Description of the photometric catalogue.

Column No. Descriptor

Part I: stellar mean properties (403,734 lines)1 Star number2 Right Ascension (J2000)3 Declination (J2000)4 Mean J-band magnitude5 Error in 〈J〉6 Mean H-band magnitude7 Error in 〈H〉8 Mean K-band magnitude9 Error in 〈K〉

10 Number of J-band measurements11 Number of H-band measurements12 Number of K-band measurements13 Variability index J

14 Kurtosis index K

15 Variability index L

16 Estimated K-band amplitudePart II: multi-epoch data (3,623,332 lines)1 Star number2 Epoch (HJD–2,450,000)3 Filter (J, H or K)4 Magnitude5 Error in magnitude

to be consistent with the results from our cross-correlationswith three other optical and IR catalogues (cf. Section 5.3).

5 DISCUSSION

5.1 The near-IR variable star population

Figure 18 presents near-IR CMDs for the whole region ofM33 monitored with WFCAM. The Large-amplitude vari-able stars we identified are highlighted in green. These aremainly found between Ks ∼ 16–18 mag, and are largelyabsent among fainter stars (below the tip of the RGB atKs ∼ 18 mag). Some brighter variable RSGs are found(around Ks ∼ 14 mag), but the clump of variables starswith Ks < 11 mag and redder colour in H−Ks than J −Ks

are foreground stars and perhaps saturated. Variable starsdominate the redder stars to the right of the vertical se-quence comprising the bulk of stars.

The distributions over brightness (Fig. 19) and colour(Fig. 20) provide another means of assessing the propertiesof the variable stars. The largest fraction of stars that arefound to be variable occurs between Ks ∼ 16–17 mag, al-beit less than that of the UIST survey in the central regionin the same magnitude interval. The variable star popula-tion reaches a peak around Ks ∼ 17 mag; it then dropsat fainter magnitudes even though the total stellar popula-tion keeps increasing. This mainly arises from two factors:firstly, many stars in this magnitude interval have not yetreached the final phase of their evolution and they will stillevolve to higher luminosities and lower temperatures be-fore they develop large-amplitude variability; secondly, thebirth-mass and K-band brightness relation flattens consider-ably for low-mas AGB stars (see Paper II). At (J −Ks) > 2mag or (H − Ks) > 1 mag almost all stars (Ks < 19 mag)

Figure 18. Near-IR CMDs (WFCAM variable stars in green).

Figure 19. Distribution of all WFCAM sources (solid) and thevariable stars (dotted), as a function of near-IR brightness.

are variable; these are dusty, strongly pulsating and heavilymass-losing AGB stars.

The level of contamination by foreground stars can beassessed with the trilegal simulation tool (Girardi et al.2005). We used default parameters for the structure of theGalaxy, simulating a 0.9 square degree field in the direction

c© 2014 RAS, MNRAS 000, 1–20

12 Javadi et al.

Figure 20. Distribution of all WFCAM sources with Ks < 19mag (solid) and the variable stars (dotted), as a function of near-IR colour.

Figure 21. Estimated contamination by foreground stars (inred), from a simulation with trilegal (Girardi et al. 2005).

Figure 22. CMD of (J − Ks), with WFCAM variable stars ingreen. Overplotted are isochrones from Marigo et al. (2008) forsolar metallicity and a distance modulus of µ = 24.9 mag.

(l = 133.61◦, b = −31.33◦). Only a relatively small numberof foreground stars are expected, with fairly neutral coloursor below our completeness limit (Fig. 21). The part of theCMD occupied by large-amplitude variable stars is relativelyuncontaminated by foreground stars.

The stellar populations can be described usingisochrones calculated by Marigo et al. (2008) (Fig. 22). Theisochrones were calculated for solar metallicity (Z⊙ = 0.015)for all stellar populations except the oldest, least chemi-cally evolved one with log t[yr] = 10 for which we adoptedZ = 0.008. To account for a (shallow) metallicity gradientacross the disc of M 33, we show CMDs of each of the 16 tilesof the M33 mosaic, overlain with isochrones for Z = 0.015in the central region but Z = 0.008 further out in the disc(Fig. 23).

The isochrones from Marigo et al. (2008) are the mostrealistic models to be used for the purpose of this study, forthe folowing reasons:

• The star’s evolution is followed all the way through thethermal pulsing AGB until the post-AGB phase. Crucially,two important phases of stellar evolution are included, viz.the third dredge-up mixing of the stellar mantle as a resultof the helium-burning phase, and the enhanced luminosity ofmassive AGB stars undergoing hot bottom burning (HBB;Iben & Renzini 1983);

• The molecular opacities which are important for thecool atmospheres of evolved stars have been considered inthe models of stellar structure. The transformation fromoxygen-dominated (M-type) AGB stars to carbon stars in

c© 2014 RAS, MNRAS 000, 1–20

M33 monitoring. IV 13

Figure 23. Near-IR CMDs of each of the individual tiles in our mosaic. The WFCAM variable stars are highlighted in green. Isochronesfrom Marigo et al. (2008) are overlain for Z = 0.015 (pink) and Z = 0.008 (blue).

c© 2014 RAS, MNRAS 000, 1–20

14 Javadi et al.

the birth mass range M ∼ 1.5–4 M⊙ is accounted for (cf.Girardi & Marigo 2007);

• The dust production in the winds of LPVs, and theassociated reddening is included;

• The radial pulsation mode is predicted;• Combination of their own models for intermediate-mass

stars (M < 7 M⊙), with Padova models for more massivestars (M > 7 M⊙; Bertelli et al. 1994), gives a completecoverage in birth mass (0.8 < M < 30 M⊙);

• Magnitudes are calculated on a wide range of commonoptical and IR photometric systems;

• The isochrones are available via an internet-based form,in a user-friendly format.

5.2 Spatial distribution of LPVs

Maps of the surface density of the number of large-amplitudevariable stars, AGB stars, RGB stars and massive stars arepresented in figure 24. The same selection criteria as in Pa-per II are used to select these different populations: thedemarcation between massive stars and less-massive giantstars is defined to run from (J −Ks, Ks) = (0.6, 18) mag to(J−Ks,Ks) = (0.9, 15.6) mag, such that massive stars havecolours bluer than this (down to Ks = 19.5 mag) or haveKs < 15.6 mag, whilst AGB stars and RGB stars are redderthan this and have 16 < Ks < 18 mag or 18.3 < Ks < 19.5mag, respectively.

The variable stars, AGB stars and massive stars are con-centrated towards the centre but the RGB stars do not showsuch strong central concentration. This is in good agreementwith what we found for the central square kpc of M33 in Pa-per II. The distribution of the variable stars mostly mimicsthat of the AGB stars, as expected, though the somewhatstronger central concentration in the former suggests thatmore massive stars (AGB stars as well as RSGs) make alarger contribution in the central part of M33 than furtherout in the disc. Only hints can be seen of the spiral armpattern in these maps.

5.3 Cross-identifications in other catalogues

We have cross-correlated our UKIRT/WFCAM variabilitysearch results with those from two intensive optical mon-itoring campaigns (CFHT, Hartman et al. 2006, in 5.3.1;DIRECT, Macri et al. 2001, in 5.3.2). We have also com-pared with the optical catalogue of Rowe et al. (2005) whichincludes narrow-band filters that they used to identify car-bon stars (in 5.3.3); RSGs from Drout et al. (2012) in 5.3.4;and a small number of Luminous Blue Variable (LBV) starsfrom Humphreys et al. (2014) in 5.3.5. In addition, we com-pared our data with the mid-IR variability search performedwith the Spitzer Space Telescope (McQuinn et al. 2007) in5.3.6. The matches were obtained by search iterations usinggrowing search radii, in steps of 0.′′1 out to 1′′, on a first-encountered first-associated basis after ordering the princi-pal photometry in order of diminishing brightness (Ks-bandfor the UKIRT catalogue, i-band/I-band for the optical cat-alogues, and 3.6-µm band for the Spitzer catalogue). Finally,we examined our data on the recently identified 24-µm vari-ables (Montiel et al. 2014) which includes the giant H ii re-gion NGC604 (in 5.3.7).

Figure 25. Near-IR CMD showing the stars from the WFCAMsurvey that were and were not identified as variable stars in theCFHT optical variability survey (Hartman et al. 2006).

5.3.1 CFHT optical variability survey

A variability survey of M33 was carried out with the 3.6-m Canada–France–Hawai’i Telescope (CFHT) on 27 nightscomprising 36 individual measurements, between August2003 and January 2005 (Hartman et al. 2006). Out of twomillion point sources in a square-degree field, they identifiedmore than 1300 candidate variable blue and red supergiants,more than 2000 Cepheids, and more than 19 000 AGB andRGB LPVs. Their catalogue comprises Sloan g′-, r′- andi′-band photometry to a depth of i′ ≈ 24 mag.

Out of 36 000 variable stars detected with CFHT in ourfield of M33, our UKIRT/WFCAM survey detected 29 600stars (82%), of which 1818 were found by us to be variable(Fig. 25). Most of the CFHT variables that were missed inour survey are fainter than the RGB tip; these are not thetype of variables that our survey aims to detect. The AGBand RSG variables that were not detected in our survey havemodest amplitudes (especially at IR wavelengths). On theother hand, our survey detected some of the dustiest AGBvariables that were missed by the CFHT survey.

5.3.2 DIRECT optical variability survey

The DIRECT variability survey (Macri et al. 2001) aimedto determine a distance estimate of M33 (and M31) usingdetached eclipsing binaries and Cepheids. It was carried outon 95 nights with the F.L. Whipple observatory’s 1.2-m tele-scope and on 36 nights with the Michigan–Dartmouth–MIT

c© 2014 RAS, MNRAS 000, 1–20

M33 monitoring. IV 15

Figure 24. Spatial distribution across M 33 of the near-IR populations of (Top left:) WFCAM LPVs; (Top right:) AGB stars; (BottomLeft:) RGB stars; (Bottom right:) massive stars. The units are logarithmic in number of stars per square kpc.

1.3-m telescope, between September 1996 and October 1997.Their catalogue contains Johnson B- and V-, and CousinsI-band photometry for all stars with 14.4 < V < 23.6 mag,and the V-band J variability index (cf. Section 3).

The central region of M 33 observed in DIRECT is com-pletely covered in our WFCAM survey. The DIRECT surveylisted 57 581 stars among which 1383 have J > 0.75 (whichthe DIRECT team assumed as the variability threshold).Within the coverage of DIRECT are located 116 606 starsfrom the WFCAM catalogue, including 1444 WFCAM vari-ables. Of the 24 422 stars from DIRECT that were detectedin our WFCAM survey, 412 were found by us to be variable.Hence, most of the WFCAM variables were missed by theDIRECT survey (Fig. 26). This is unsurprising since dustyAGB variables are faint at optical wavelengths, but the rel-atively short duration of the DIRECT survey (little more

than a year) may have contributed to it missing also lessdusty AGB variables.

5.3.3 Carbon star survey

Rowe et al. (2005) used the CFHT and a four-filter sys-tem in 1999 and 2000 to cover much of M33. The filtersystem was designed to identify carbon stars on the basisof their cyanide (CN) absorption as opposed to other redgiants that display titanium-oxide (TiO) absorption, usingnarrow-band filters centered on 8120 and 7777 A, respec-tively. They added broad-band Mould V- and I-band filtersto aid in selecting cool stars. Carbon stars in their schemehave [CN]−[TiO] > 0.3 and V − I > 1.8 mag, and M-typestars have [CN]−[TiO] < −0.2 mag at the same V–I crite-rion (they only considered stars for this purpose that haderrors on these colours of < 0.05 mag).

c© 2014 RAS, MNRAS 000, 1–20

16 Javadi et al.

Figure 26. As Fig. 25, for the DIRECT optical variability survey(Macri et al. 2001).

Within the area in common with our WFCAM survey,Rowe et al. detected 2 079 334 stars, from which 306 292 stars(3.5%) were identified with WFCAM (a further 85 098 starsfrom the WFCAM catalogue are outside of their coverage).Among these, 11 040 stars are M-type and 3134 are car-bon stars. To improve the cross-correlation and to avoid co-incidences because of the high density of optical sources,a pre-selection was made on carbon stars and M-type starsthat are likely to have been detected at near-IR wavelengths.The average offset between the two catalogues was found tobe only 0.′′02 in both RA and DEC.

The main giant branches – viz. the RSGs around(J − Ks) ≈ 0.8 mag and the AGB around (J − Ks) ≈ 1mag – are traced by M-type stars, with carbon stars di-gressing towards redder colours at Ks ∼ 16–17 mag (Fig.27). The carbon star distribution is consistent with a typi-cal t ∼ 1 Gyr (log t ∼ 9) isochrone or a little younger, i.e.birth masses around 2–3 M⊙. Some of the bright carbonstars show signs of reddening to (J − Ks) > 2 mag pre-sumably due to circumstellar dust; as these are among thebrightest detected carbon stars, they must indicate the ter-mination point in carbon star evolution. Carbon stars arefound fainter than the RGB tip, down to Ks ∼ 19 mag;these may have formed through binary mass transfer. Suchfaint carbon stars are known in other, especially metal poorpopulations, e.g., in the Sagittarius dwarf irregular galaxy(Gullieuszik et al. 2007) or in the Galactic globular clusterωCentauri (van Loon et al. 2007). Note that our WFCAMsurvey detected large-amplitude variability in carbon starsonly brighter than Ks ≈ 17.5 mag, i.e. above the RGB tipand consistent with thermal pulsing AGB stars that havebecome carbon stars as a result of third dredge-up.

Figure 28 presents a near-IR colour–colour diagram forAGB and RGB stars only, as other populations tend to con-fuse the picture in such diagram. Stars with 18.5 < Ks <19.5 mag are assumed to be on the RGB, whilst stars with

Figure 27. Near-IR CMD showing the M-type and carbon starsfrom the Rowe et al. (2005) survey that were detected in ourUKIRT/WFCAM survey. A 1-Gyr isochrone from Marigo et al.(2008), for solar metallicity, is shown for comparison.

16 < Ks < 18 mag are considered to be on the AGB (notethat this selection excludes some of either type but max-imises the purity of these two samples). For both selectionswe only kept stars with photometric errors on the colour< 0.25 mag. The M-type and carbon stars classified by Roweet al. are highlighted. Some of the M-type, but especially car-bon stars follow part of a sequence towards red colours, butthe reddest of these, at (H −Ks) > 1 mag and (J −H) > 1mag have generally not been bright enough at optical wave-lengths for spectral typing. These are dusty LPVs identifiedin our WFCAM survey. The realm of the RGB stars spreadsout over colour but generally this happens in one but notboth colours at once, suggesting blends and/or other photo-metric uncertainties are to be blamed.

5.3.4 Red Supergiant stars

Drout et al. (2012) identified RSGs (and yellow supergiants)in M33 using the Hectospec multi-fiber spectrograph onthe 6.5-m Multiple Mirror Telescope in two observing cam-paigns, in 2009 and 2010.

They divided their list in three different categories:those supergiants that were selected both photometricallyand kinematically, assigned rank 1; those confirmed withjust one method, assigned rank 2; and those which were notselected by either method and which are likely foregrounddwarfs, assigned rank 3. They identified 189 rank-1 stars, 12rank-2 stars and 207 rank-3 stars.

c© 2014 RAS, MNRAS 000, 1–20

M33 monitoring. IV 17

Figure 28. Near-IR colour–colour diagram showing the M-typeand carbon stars from the Rowe et al. (2005) survey that weredetected in our UKIRT/WFCAM survey. Stars with 18.5 < Ks <

19.5 mag and errors on the colours < 0.25 mag are labeled asRGB stars, whilst stars with 16 < Ks < 18 mag and errors onthe colours < 0.25 mag are labeled as AGB stars.

Figure 29. As Fig. 26, for the RSGs from Drout et al. (2012).

Our WFCAM survey detected 381 of the red stars inthe survey by Drout et al. (93%), from which 186 rank-1stars (98%) and 13 rank-2 stars. Of the rank-1 stars, 14were found by us to be variable, as was one rank-2 star.Their RSGs clearly delineate a branch in the near-IR CMD(Fig. 29; see also Fig. 30). It is possible that some of thereddened sources are also RSGs but they will have been too

Figure 30. Light curves of selected variable stars: (Top:) theLBV Var C; (Middle:) two of the RSGs; (Bottom:) one of thestars for which we showed the lightcurve in Paper I, but nowbased on the combined UIST+WFCAM photometry.

faint for spectral typing and hence not been included in thework by Drout et al. (2012). Three of the variables in thetop of the AGB branch may instead be massive AGB starsor super-AGB stars (Fig. 29).

5.3.5 The Luminous Blue Variable Var C

Humphreys et al. (2014) identified a small number of LBVsin M33, one of which is called “Var C”. Since its discovery,a series of eruptions have been witnessed in this star: 1940–1953 (Hubble & Sandage 1953); 1964–1970 (Rosino & Bian-chini 1973); and 1982–1993 (Humphreys et al. 1988; Szeifertet al. 1996). By 1998 it had returned to a minimum state(Burggraf et al. 2014), but soon another, shorter episode ofmaximum light was seen from 2001 until 2005 (Viotti et al.2006; Clark et al. 2012). Currently it is in a hot quiescentstage (Humphreys et al. 2014).

Var C was detected in our WFCAM survey four, fiveand ten times in the J-, H- and K-band, respectively. Thephotometry is reliable since there is no neighbouring starbright enough to seriously affect the photometry. The meanmagnitudes are 〈J〉 = 16.32 mag, 〈H1〉 = 6.14 mag and〈Ks〉 = 15.75 mag; with 〈J − Ks〉 = 0.57 mag it is one ofthe bluer confirmed variable stars in our list. Over the twoyears of our monitoring campaign with WFCAM, Var C hassteadily diminished its brightness, though in 2007 it seemsto have stabilised (Fig. 30).

5.3.6 Spitzer mid-IR variability survey

Five epochs of Spitzer Space Telescope imagery in the 3.6-,4.5- and 8-µm bands have been analysed by McQuinn et al.(2007), to identify variable stars using a smilar method tothat we used ourselves.

The Spitzer images covered nearly a square degree,

c© 2014 RAS, MNRAS 000, 1–20

18 Javadi et al.

Figure 31. As Fig. 25, for the Spitzer mid-IR variability survey(McQuinn et al. 2007).

slightly larger than our WFCAM survey; out of 40 571Spitzer sources, 2868 stars fall outside theWFCAMmonitor-ing coverage. Among the stars that Spitzer detected withinthe region in common with our survey, 36 411 are also inour photometric catalogue, down to a little below the RGBtip (Fig. 31). Hence, the recovery rate is ∼ 97%. The recov-ery rate of the Spitzer survey for RSGs, bright AGB starsand dusty AGB star variables from our WFCAM survey isgood. Blending is only problematic for stars within the cen-tral square kpc of M33 (cf. Paper I), where the recovery ratedecreases to ∼ 70%.

Among the 2923 variable stars identified with Spitzer,985 were identified as variable stars also in our WFCAMsurvey. This corresponds to a recovery rate of 35% (exclud-ing 113 stars that fall outside the WFCAM coverage), i.e.slightly higher than the recovery rate by the WFCAM sur-vey of UIST variable stars in the central square kpc (cf.Section 3.1). On the other hand, 3661 of our WFCAM vari-able stars had not been identified as variable stars in theSpitzer survey. Both surveys do well in detecting variabledusty AGB stars; the WFCAM survey is also sensitive tofainter, less dusty variable red giants (Fig. 32) that were toofaint for Spitzer.

Figure 33 shows a mid-IR CMD, with a well-populatedsequence off from which a branch extends towards reddercolours and fainter 3.6-µm brightness starting around [3.6] ∼14–16 mag and ([3.6]− [4.5]) > 0.2 mag. This branch mostlycomprises dust-enshrouded objects, with a high fraction ofWFCAM variable stars. The WFCAM variables are plentifulalso among the brighter 3.6-µm sources in the diagram; these

Figure 32. The variable stars that were or were not identifiedwith WFCAM or Spitzer.

Figure 33. Mid-IR CMD of Spitzer photometry of our UKIRT/WFCAM sources, with WFCAM variables highlighted in green

and M-type and carbon stars from Rowe et al. (2005) in blue andred, respectively. Isochrones from Marigo et al. (2008) for 10 Myr,100 Myr and 1 Gyr are overlain for comparison.

are massive AGB stars and RSGs. The M-type stars form asequence of increasing 3.6-µm brightness, with carbon starslocated along part of this sequence. The near absence of M-type or carbon stars in the red branch of dust-enshroudedobjects is due to the limited sensitivity of the optical spectraltyping survey. The Padova isochrones seem to underestimatethe 3.6-µm brightness by a factor three or so, most notablein the missing of the 1-Gyr isochrone of the red branch.

c© 2014 RAS, MNRAS 000, 1–20

M33 monitoring. IV 19

Figure 34. Near-IR CMD showing the 24-µm variables fromMontiel et al. (2014) that were detected in our UKIRT/WFCAMsurvey. The variability indication refers to our near-IR survey.

5.3.7 Sources variable at 24 µm

Montiel et al. (2014) have identified 24 sources in M33 thatare variable at a wavelength of 24 µm, based on Spitzerdata with the MIPS instrument. One of these is NGC604(see below); the other sources could be dusty evolved stars.

We find that VC6, 8, 9, 13, 16 21 and 23 (their nomen-clature) are variable with high confidence and VC10, 17 and20 probably (but with fewer near-IR epochs so less reli-able). These are all red, with (J − Ks) > 2.4 mag (Fig.34). We do not detect variability in VC14 but its photome-try – (J −Ks) = 0.94 mag, Ks = 13.32 mag – is consistentwith being a RSG and thus confirms the identification byMontiel et al. with the M2-type 760-d semiregular variablestar VHK71 (van den Bergh, Herbst & Kowal 1975); VC17is likely variable and its photometry – (J −Ks) = 3.5 mag,Ks = 14.87 mag – is consistent wiht it being a dusty RSG.

On the other hand, for VC7 we only find a moderatelyred star without detected variability; this is probably thecounterpart of the optical variable star with a 143-d periodwith which Montiel et al. associated VC7, but not of the24-µm variable itself as the 10-µm absorption in the latterappears incompatible with such rapid variability.

None of the 24-µm variables with ([3.6] − [8.0]) > 4mag – which includes all of their far-IR detections – wereidentified by us as near-IR variables. If this is because theyare in fact not dusty evolved stars then this would explainwhy they do not obey the evolved stars sequences in their[3.6]–[8.0] vs. [3.6]–[4.5] diagram; they may instead be youngstellar objects.

Figure 35. WFCAM JHKs composite of the NGC 604 H ii regionin M 33. The WFCAM variable stars are identified in red.

VC1 in Montiel et al. (2014) corresponds to NGC604,the second largest extra-galactic H ii region in the LocalGroup (after 30Doradus in the LMC). It is a young starforming region with an age of 3–5 Myr (e.g., Wilson &Matthews 1995; Pellerin 2006). Located some 12′ from thecentre of M 33, it spans around 4.1 pc with a core–halo struc-ture (Melnick 1980). NGC604 has been studied throughoutthe electromagnetic spectrum, including radio (Churchwell& Grass 1999; Tosaki et al. 2007), infrared (Higdon et al.2003), optical (Tenorio-Tagle et al. 2004), ultraviolet (Keelet al. 2004), and X-ray (Maız-Apellaniz et al. 2004). Thesestudies were mostly concerned with the effect of the massivestar formation on the surrounding ISM. Using WFCAM wehave identified 12 near-IR variable stars within NGC604,which are shown in figure 35.

In the near-IR CMD (Fig. 34), the sources withinNGC604 are neither particularly luminous nor red – in fact,most have (J − Ks) < 1 mag. We suspect that this groupof variables may include any or all of the following types ofsources: (i) young early-type stars; (ii) stars whose photom-etry is affected by nearby stars and/or nebulosity; and (iii)a dusty carbon star not born in NGC604 (at (J−Ks) = 1.7mag).

6 CONCLUSIONS

WFCAM on UKIRT was used to extend our near-IR mon-itoring survey of the Local Group spiral galaxy M33, fromthe central square kpc to a square degree. K-band obser-vations were complemented with occasional J- and H-bandobservations to provide colour information.

The photometric catalogue comprises 403 734 stars,among which 4643 stars display large-amplitude variability.Investigation of the lightcurves and location on the near-IRCMD with respect to theoretical models of stellar evolutionindicate that most of these stars are AGB stars or RSGs.They are concentrated towards the centre of M 33, more so

c© 2014 RAS, MNRAS 000, 1–20

20 Javadi et al.

than RGB stars. The majority of very dusty stars which areheavily reddened even at IR wavelengths are variable.

Cross-matching with optical monitoring campaigns andmid-IR variability searches conducted with Spitzer, showsthat the UKIRT/WFCAM catalogue of variable stars isvastly more complete for the dusty variables than the opticalsurveys, and more complete for less dusty variables than theSpitzer surveys. Our catalogue is made publicly available atthe CDS.

This work forms the basis for the next two papers inthis series, to derive the SFH (Paper V) and to measure therate and location of dust production (Paper VI) across M33.

ACKNOWLEDGMENTS

We thank the staff at UKIRT for their excellent supportof this programme, and the referee for her/his constructivereport. JvL thanks the School of Astronomy at IPM, Tehran,for their hospitality during his visits. We are grateful forfinancial support by The Leverhulme Trust under grant No.RF/4/RFG/2007/0297, by the Royal Astronomical Society,and by the Royal Society under grant No. IE130487.

REFERENCES

Benjamin R. A., et al., 2005, ApJ, 630, L149

Bonanos A. Z. et al., 2006, ApJ, 652, 313Burggraf B., et al., 2014, A&A, submitted

Clark J. S., et al., 2012, A&A, 541, A146

Churchwell E., & Goss W. M., 1999, ApJ, 514, 188

Cioni M.-R. L., et al., 2008, A&A, 487, 131Deul E. R., van der Hulst J. M., 1987, A&AS, 67, 509

Drout M., Massey P., Meynet G., 2012, ApJ, 750, 97

Engargiola G., Plambeck R. L., Rosolowsky E., Blitz L., 2003,ApJ, 149, 343

Girardi L., Groenewegen M. A. T., Hatziminaoglou E., da CostaL., 2005, A&A, 436, 895

Gullieuszik M., Rejkuba M., Cioni M. R., Habing H. J., Held E.V., 2007, A&A, 475, 467

Hartman J. D., Bersier D., Stanek K. Z., Beaulieu J.-P., Ka luznyJ., Marquette J.-B., Stetson P. B., Schwarzenberg-Czerny A.,2006, MNRAS, 371, 1405

Higdon S. J. U., Higdon J. L., van der Hulst J. M., Stacey G. J.,2003, ApJ, 592, 161

Hodierna G. B., 1654, De Systemate Orbis Cometici, Deque Ad-mirandis Coeli Characteribus (About the systematics of thecometary orbit, and about the admirable objects of the sky),Palermo

Hubble E., Sandage A., 1953, ApJ, 118, 353

Humphreys R. M., Leitherer C., Stahl O., Wolf B., Zickgraf F.-J.,1988, A&A, 203, 306

Iben I. Jr., Renzini A., 1983, ARA&A, 21, 271

Javadi A., van Loon J. Th., Mirtorabi M. T., 2011a, MNRAS,411, 263 (Paper I)

Javadi A., van Loon J. Th., Mirtorabi M. T., 2011b, MNRAS,414, 3394 (Paper II)

Javadi A., van Loon J. Th., Mirtorabi M. T., 2011c, in: WhyGalaxies Care About AGB Stars II, eds. F. Kerschbaum, T.Lebzelter & R. F. Wing, ASPC, 445, 497

Javadi A., van Loon J. Th., Khosroshahi H., Mirtorabi M. T.,2013, MNRAS, 432, 2824 (Paper III)

Keel W. C., Holberg J. B., Treuthardt P. M., 2004, AJ, 128, 211

Levesque E. M., 2010, in: Hot and Cool – Bridging Gaps in Mas-sive Star Evolution, eds. C. Leitherer, P. Bennett, P. Morrisand J. Th. van Loon, ASPC (San Francisco: ASP), 425, p103

Levesque E. M., Massey P., Olsen K. A. G., Plez B., Josselin E.,Maeder A., Meynet G., 2005, ApJ, 628, 973

Macri L. M., Stanek K. Z., Sasselov D. D., Krockenberger M.,

Ka luzny J., 2001, AJ, 121, 861Maız-Apellaniz J., Perez E., Mas-Hesse J. M., 2004, AJ, 128, 1196Marigo P., Girardi L., Bressan A., Groenewegen M. A. T., Silva

L., Granato G. L., 2008, A&A, 482, 883Massey P., Olsen K. A. G., Hodge, P. W., Strong S. B., Jacoby

G. H., Schlingman W., Smith R. C., 2006, AJ, 131, 2478Melnick J., 1980, A&A, 86, 304McQuinn K. B. W., et al., 2007, ApJ, 664, 850Messier C., 1771, Mem. Acad., p448Mochejska B. J., Ka luzny J., Stanek K. Z., Sasselov D. D., Szent-

gyorgyi A. H., 2001a, AJ, 121, 2032Mochejska B. J., Ka luzny J., Stanek K. Z., Sasselov D. D., Szent-

gyorgyi A. H., 2001b, AJ, 122, 2477Montiel E. J., Srinivasan S., Clayton G. C., Engelbracht C. W.,

Johnson C. B., AJ, in press (arXiv:1411.6008)Pellerin A., 2006, AJ, 131, 849Pierce M. J., Jurcevic J. S., Crabtree D., 2000, MNRAS, 313, 271Pietsch W., Misanovic Z., Haberl F., Hatzidimitriou D., Ehle M.,

Trinchieri G., 2004, A&A, 426, 11Rosino L., Bianchini A., 1973, A&A, 22, 453Rosolowski E., Engargiola G., Plambeck R., Blitz L., 2003, ApJ,

599, 258Skrutskie M. F., et al., 2006, AJ, 131, 1163Stetson P. B., 1987, PASP, 99, 191Stetson P. B., 1993, in: Stellar Photometry – Current Techniques

and Future Developments, eds. C. J. Butler and I. Elliott, IAUColl. Ser. 136 (Cambridge: Cambridge University Press), p291

Stetson P. B., 1996, PASP, 108, 851Szeifert T., Humphreys R. M., Davidson K., Jones T. J., Stahl

O., Wolf B., Zickgraf F.-J., 1996, A&A, 314, 131Rowe J. F., Richer H. B., Brewer J. P., Crabtree D. R., 2005, AJ,

129, 729Tenorio-Tagle G., Munoz-Tunon C., Perez E., Maız-Apellaniz J.,

Medina-Tanco G., 2000, ApJ, 541, 720Thompson T. A., Prieto J. L., Stanek K. Z., Kistler M. D., Bea-

com J. F., Kochanek C. S., 2009, ApJ, 705, 1364Tosaki T., Miura R., Sawada T., Kuno N., Nakanishi K., Kohno

K., Okumura S. K., Kawabe R., 2007, ApJ, 664, L27van den Bergh S., Herbst E., Kowal C. T., 1975, ApJS, 29, 303van Loon J. Th., 2010, in: Hot and Cool – Bridging Gaps in Mas-

sive Star Evolution, eds. C. Leitherer, P. Bennett, P. Morrisand J. Th. van Loon, ASPC (San Francisco: ASP), 425, p279

van Loon J. Th., Zijlstra A. A., Whitelock P. A., Waters L. B. F.M., Loup C., Trams N. R., 1997, A&A, 325, 585

van Loon J. Th., et al., 1999, A&A, 351, 559van Loon J. Th., Cioni M.-R. L., Zijlstra A. A., Loup C., 2005,

A&A, 438, 273van Loon J. Th., van Leeuwen F., Smalley B., Smith A. W., Lyons

N. A., McDonald I., Boyer M. L., 2007, MNRAS, 382, 1353van Loon J. Th., Cohen M., Oliveira J. M., Matsuura M., McDon-

ald I., Sloan G. C., Wood P. R., Zijlstra A. A., 2008, A&A,487, 1055

Viotti R. F., Rossi C., Polcaro V. F., et al., 2006, A&A, 458, 225Whitelock P. A., Feast M. W., van Loon J. Th., Zijlstra A. A.,

2003, MNRAS, 342, 86Wilson C. D., Matthews B. C., 1995, ApJ, 455, 125Wood P. R., 1998, A&A, 338, 592Wood P. R., Whiteoak J. B., Hughes S. M. G., Bessell M. S.,

Gardner F. F., Hyland A. R., 1992, ApJ, 397, 552Zaritsky D., Elston R., Hill J. M., 1989, AJ, 97, 97

c© 2014 RAS, MNRAS 000, 1–20